The Pentagon has moved at least two-thirds of its AI workload away from Anthropic and toward rival providers, marking one of the most dramatic vendor shake-ups in the Defense Department’s short but intense history with commercial AI.
The shift follows a months-long dispute between the DoD and Anthropic over acceptable use policies for Claude, the company’s flagship AI model. What started as a contract negotiation escalated into a full-blown standoff over where to draw the line between AI safety principles and national security demands.
How the rift unfolded
The backstory starts in July 2025, when the Pentagon awarded $200 million contracts each to Anthropic, OpenAI, Google, and xAI for AI prototyping work.
Things started to unravel in early 2026. The Pentagon pushed Anthropic to permit its models for unrestricted lawful uses, a category that reportedly included mass surveillance capabilities and autonomous weapons deployment without human oversight. Anthropic CEO Dario Amodei declined.
By February 2026, the dispute had reached its peak. The Pentagon wasn’t interested in philosophical debates about AI alignment. It wanted tools that worked without usage restrictions, and it wanted vendors who wouldn’t push back on how those tools got deployed.
In March 2026, the Defense Department took the extraordinary step of formally designating Anthropic as a supply-chain risk, effectively putting Anthropic on a temporary federal blacklist, though a judge later blocked that move on First Amendment grounds.
Parts of the original $200 million contract were cancelled outright.
Who benefits from Anthropic’s loss
By June 2026, the DoD had formalized contracts with OpenAI, Google, and Microsoft, building out what officials described as a multi-vendor strategy for AI deployment across classified networks.
OpenAI moved fastest. The company announced expanded access to its services shortly after the Anthropic rift became public. OpenAI has been steadily building its defense portfolio, and the Anthropic fallout handed it a significant opening.
Google and Microsoft, both of which already maintain deep relationships with the federal government through cloud computing contracts, were natural beneficiaries as well.
The bigger picture for AI governance
OpenAI quietly updated its usage policies in early 2024 to remove a blanket prohibition on military applications, and it has been building government-focused products ever since. That strategic bet is now paying tangible dividends in the form of expanded Pentagon contracts.
What this means for investors
The Pentagon’s pivot has immediate implications for anyone tracking the AI competitive landscape. OpenAI, still private but reportedly exploring various paths toward liquidity, just added significant government revenue to its profile.
Investors should watch whether the supply-chain risk designation against Anthropic gets permanently struck down or merely paused. If the legal precedent holds that the government can’t blacklist AI companies for refusing specific use cases, it establishes a meaningful protection for the entire industry. If it doesn’t, every AI company’s acceptable use policy becomes a potential liability in government contract negotiations.